recovery_tests: Mamba2SSM forward_with_gradients+backward+optimizer_step gpu_kernel_parity: collect_experiences_gpu, store() not vars(), CudaSlice readback dqn_gradient_collapse: GpuTensor::randn+to_dtype, host-side gather dqn_diagnostic: GpuTensor::from_host, to_host for normalization check trainable_adapter: MlDevice import gated #[cfg(test)] Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
17 KiB
17 KiB